← Back to Posts

Reliable Sparse Representation Learning

Heteroscedastic Orthogonal Transform Learning via Soft Consistency

How should a sparse model learn when some observations are more reliable than others? We use local observation uncertainty and model error to guide transform learning, sparse coding, and reconstruction. This video introduces the intuition, optimization objective, and calibration workflow, with demonstrations of Gaussian denoising and an extension to masked inpainting.

English narration · 4:26 · Open video · Download MP4